You can use SAR data and LiDAR as done in the following study:

https://pdfs.semanticscholar.org/c0e5/9713c574a7c2634bb55893193d98fe4dd69d.pdf

The BREC toll may help: Conference Paper BREC: The built-up area RECognition tool

There is a version of BREC at InterIMAGE (http://www.lvc.ele.puc-rio.br/projects/interimage/), GEOBIA brazilian open GIS, its the BREC Urban Focus.

Hi Marcos, I will have a look.

Many thanks!

Alessandro

Dear Sita, thanks for your suggestion.

At the moment I am looking to define the footprints from multispectral data only.

Alessandro

Hi Melanie, many thanks for your input!

Cheers,

Alessandro

Alessandro, I saw the following GGIS toolbox being recommended in a similar question

http://www.rasor.eu/wiki/index.php/RASOR_QGIS_plugins

I don't know it personally though. have a look.

Good luck, Norman

Dear Matthieu and Norman, many thanks for your answers!

Alessandro

Dear Alessandro,

If you look at the outcomes of the different SpaceNet challenges you will find relevant information too. For example there are nicely written blog post from the winners https://towardsdatascience.com/mapping-challenge-winning-solution-1aa1a13161b3 https://medium.com/the-downlinq/2nd-spacenet-competition-winners-code-release-c7473eea7c11 https://medium.com/the-downlinq/the-spacenet-challenge-off-nadir-buildings-introducing-the-winners-b60f2b700266

What is interesting is to see that sometimes simpler architectures like U-NET are beating the more complex ones such as Mask-RCNN!

In the end, the training data representativity matters most and ensemble of networks and post-processing are giving a boost.

You can try out the hosted model for building footprint on our platform (https://picterra.ch) or directly retrain one there too.

Best,

Frank

Hi Frank de Morsier thanks! I will do.

You can use GIS Software depending on your need. ArcGIS and Envi have powerful tools